Fullscreen HTML5Med Res (??MB)HTML4

A human in the loop Techno Talk HITL used to stand for hardware-in-the-loop (eg, hardware as part of a simulation). Now, HITL also stands for human-in-the-loop, where human judgement is added to automated systems. I’m rarely exposed to something new in isolation. Usually, it doesn’t take long before I’m up to my ears in more instances or variants of the same thing. Take robots, for example. In my previous column, we considered some of the latest and greatest developments in humanoid robots. As we noted in that column, human environments—like homes, offices and factories—are designed for humans. As a result, robots with legs are better than their wheeled counterparts for navigating things like stairs. Similarly, robots with humanoid arms and hands can employ tools, operate machines and manipulate objects that were intended for use by people. It’s also true that when robots look more like people, they can interact with humans more naturally, using familiar mannerisms (gestures, movements and motions). This facilitates their use in roles like companions, caregivers, receptionists and tutors; cases where human-like interaction will dramatically enhance the user experience (or is intrinsic to the role). First, the good news Robots have been around for quite some time. The first successful industrial robot in the modern sense was arguably the Unimate from Unimation Inc, which was first deployed in a General Motors factory in 1961. The Unimate was an industrial robot arm, not a humanoid robot, but it marked the first step (no pun intended) toward real-world robotics as we think of it today. Today’s robots are a world away from the Unimate. Some very impressive examples of humanoid robots are: • Apollo from Apptronik (https://youtu.be/pymvNott6nw) • Atlas from Boston Dynamics (https://youtu.be/I44_zbEwz_w) • Digit from Agility Robotics (https://youtu.be/0dexTZ7r02Q) • Optimus from Tesla (https://youtu.be/jQd128YTtQQ) • NEO Gamma from 1X (https://youtu.be/uVcBa6NXAbk) Now for the bad news These examples are on the cutting 24 edge of humanoid robotics. They can walk, balance, carry objects, climb stairs and perform basic manual tasks (sometimes with surprising grace). However, they also share several attributes and limitations that currently make them more futuristic marvels than everyday tools. For example, they rely on advanced, heavy-duty artificial intelligence (AI) for perception, decision-making, balance, navigation and manipulation tasks. These AIs are typically powered by artificial neural networks (ANNs), reinforcement learning and large-scale training datasets. In many cases, they depend on cloudbased assistance for complex tasks or learning. Despite their robust appearance, these robots are sensitive to falls, environmental challenges and unpredictable variables. So regular maintenance, calibration and software updates are essential. Their battery life is often measured in hours, not days. Charging and power management are still major concerns. Also, there are legal, ethical and safety concerns with robots interacting with humans in public or private spaces. Then there’s the cost. Estimates vary, but these robots can be in the hundreds of thousands of dollars range. Even Tesla’s Optimus, which might aim for a $20–30k (£15-22.5k) price point, is still far from affordable by the average person. Max the Magnificent only likely to get worse for businesses looking for staff. Perhaps this explains why Markets and Markets are predicting that the humanoid robot industry will exceed $17 billion by 2027, growing at a compounded annual rate of more than 50% from 2022. Meet TASKBOTs All of this leads us to TASKBOTs from REVOBOTS. These beauties can be fixed or mobile with one or multiple arms, where the arms can handle a 150-pound (70kg) payload. Their end-effectors (hands) can have sub-micron accuracy. One of the TASKBOT’s claims to fame is that its structural components are 3D-printed. In fact, REVOBOTS’ CTO, Kent Gilson, tells me that the entire body of a TASKBOT can be Feeling our age Developed countries like the UK are facing the problem of ageing populations. According to the folks at REVOBOTS (revobots.ai), by 2030, shrinking labour forces could leave over 85 million jobs unfilled, resulting in $8.5 trillion in lost annual revenue if nothing is done. High infl­ation and stagnant wages will compound the difficulty; the situation is Practical Electronics | July | 2025 Techno Talk Max the Magnificent produced using only around $1000 (£750) worth of raw materials. That sounds good to me! TASKBOTs are equally parsimonious on the computational front. Most of today’s humanoid robots are controlled by large language models (LLMs) that demand expensive, highend, power-guzzling processors. By comparison, the TASKBOT employs a set of small language models (SLMs) that are combined hierarchically, can be trained quickly and have a very small memory footprint. In fact, TASKBOT’s computational requirements can be satisfied by a $50 dual-core ZYNQ SoC (system-on-achip) FPGA (field-programmable gate array) from Xilinx/AMD in each arm, plus an Arduino to control the track subsystem in the base. A human in the loop The HITL initialism traditionally refers to a test method whereby real hardware components are integrated into a simulation or emulation environment to check their behaviour in a controlled setting. More recently, HITL has been adopted to mean human-in-the-loop, where humans actively participate in the training, evaluation, or operation of AI systems. HITL systems can learn and adapt to new situations more effectively by incorporating human feedback. Picture the scene—a new human worker turns up at a factory to work on a production line. The supervisor will show the locations of the restroom and canteen facilities and demonstrate what is required to do the job. Now, imagine a TASKBOT arriving in the same way. In this case, a remotely located human would use telepresence to control the robot. The supervisor would first guide the TASKBOT/ human combo to the battery bank (like taking a coffee break, the TASKBOT can change its own batteries when necessary). The supervisor would then demonstrate the job with the HITL replicating the actions (including recognising objects and manipulating them as required). At the same time, the TASKBOT’s sensors would be streaming everything to the cloud, which would then train the SLMs that would be downloaded into the TASKBOT’s processors. After that, the HITL would close the connection and then leave the TASKBOT to perform its duties. I tell you, every day the world gets weirder and more wonderful at the same time. I don’t know about you, but I’m excited about what these robots could do for us in the PE near future! FIND ALL YOUR ELECTRONIC COMPONENTS IN ONE PLACE BASIC MICRO E L E CT R O N I C S C O M P O N E N T S U P P L I E R w w w . basicmicro . co . u k High-quality, genuine parts Practical Electronics | July | 2025 25